Quality-Dependent Stochastic Networks: Is FIFO Always ...
Transcript of Quality-Dependent Stochastic Networks: Is FIFO Always ...
Unpublishedworkingdraft.
Notfordistribution.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
Quality-Dependent Stochastic Networks:Is FIFO Always Better Than LIFO?
Yair Liberman∗
Department of Statistics and Operations Research,
School of Mathematical Sciences
Tel Aviv University, Israel
Uri Yechiali∗
Department of Statistics and Operations Research,
School of Mathematical Sciences
Tel Aviv University, Israel
ABSTRACTIn various service and production systems, such as metal process-
ing, fresh food industry, polymer forming, etc., the quality of the
final product is a function of its total sojourn time in the system.
We consider (i) an M/G/1-type single-server queue, as well as
(ii) open tandem Jackson networks, and analyze the quality of
products traversing through each system under two service dis-
ciplines â FIFO and LIFO. Although the mean sojourn times in
anM/G/1 queue are equal under both service disciplines (that is,
E[WLI FO ] = E[WF I FO ]), the variance ofWLI FO is larger than the
variance ofWF I FO . We show that, when quality is the measure of
performance, FIFO is not necessarily better than LIFO. We consider
several service-time distributions and show under what values of
the parameters one discipline is better (or worse) than the other.
In particular, the mean quality under LIFO is better than the mean
quality under FIFO for all values of the traffic intensity. Moreover,
the mean quality under the FIFO service regime drops sharply to
0 when the traffic intensity approaches 1, while the mean quality
under LIFO is bounded below. Numerical results as functions of the
system parameters are presented and discussed for both theM/G/1queue and the tandem Jackson network.
CCS CONCEPTS•Mathematics of computing→Queueing theory; •Networks→ Network performance evaluation.
KEYWORDSQuality, FIFO, LIFOM/G/1,Tandem Jackson Network
ACM Reference Format:Yair Liberman and Uri Yechiali. 2019. Quality-Dependent Stochastic Net-
works: Is FIFO Always Better Than LIFO?. In Proceedings of ValueTools ’20:13th EAI International Conference on Performance Evaluation Methodolo-gies and Tools (ValueTools ’20). ACM, New York, NY, USA, 8 pages. https:
//doi.org/10.1145/1122445.1122456
∗Both authors contributed equally to this research.
Unpublished working draft. Not for distribution.Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for profit or commercial advantage and that copies bear this notice and the full citation
on the first page. Copyrights for components of this work owned by others than ACM
must be honored. Abstracting with credit is permitted. To copy otherwise, or republish,
to post on servers or to redistribute to lists, requires prior specific permission and/or a
fee. Request permissions from [email protected].
ValueTools ’20, May 18-20, 2020, Tsukuba, Japan© 2019 Association for Computing Machinery.
ACM ISBN 978-1-4503-9999-9/18/06. . . $15.00
https://doi.org/10.1145/1122445.1122456
1 INTRODUCTIONQueueing theory deals primarily with probabilistic analyses of
various types of service systems under a variety of probabilistic
assumptions on the processes driving such systems. Usually the
study is directed towards analyzing the distribution of the number
of customers in the system and of their sojourn times, as well as
finding the corresponding means and variances. In many produc-
tion systems, e.g. fresh food industry, metal processing, polymer
forming, etc., the quality of products traversing through the system
deteriorates along with their sojourn time (see e.g. Liberopoulos et
al, 2007 [18]). In such cases it is essential to determine the order of
service of products, namely, serving them according to the First In
First Out (FIFO) regime, or alternatively according to the Last In
Last Out (LIFO) regime (Other services regimes, such as Random
Order Of Service or Processor Sharing are not discussed in the
current study). Also, it is important to determine the best order in
which the product passes through a network of service sites, and
as a result improve the product’s quality and thus its value and
reputation (see, e.g., a survey by Inman et al., 2013 [9]).
Deterioration of products is treated extensively in the manufac-
turing and inventory management literature, and it often aims at
calculating the level of inventory stock and necessary lead time in
purpose of minimizing costs and avoiding shortage (see e.g. Van
Horenbeek et al., 2013 [25]). The inventory’s quality analysis mainly
deals with defective items due to imperfect manufacturing process,
or damages (see e,g. see Kim and Gershwin, 2005 [14]). In addition,
there is an abundance of research over quality-dependent prod-
ucts, where the majority of works deal with perishable products,
with quality defined as a time-based dichotomous function (see e.g.,
Cooper, 2001 [6]; Berk and Gurler, 2008 [2]; Liberopoulos et al., 2007
[19]; Avinadav and Arponen, 2009 [1]). Naebulharam and Zhang
(2013) [20] studied a Bernoulli quality model which determines the
quality of each part, defective or non-defective, by a series of i.i.d.
Bernoulli random variables. Coledani et al. (2015) [5] considered a
discrete-time discrete-state Markov chain to examine the effects of
buffer sizes on the probability that a product is defective or not in a
multi-stage production system. Bortolini et al. (2016) [3] described
a piecewise linear function between quality and time, and used it to
estimate the market purchase probability. Goyal and Giri (2001) [8]
and Perlman and Yechiali (2019) [21] assumed an exponential decay
function of the product’s quality. Our model also follows the expo-
nential decay assumption and quantitatively defines the quality of
a product passing through a production (or service) system.
The sojourn time of a product depends on various system char-
acteristics, such as job-arrival process, distribution of service (pro-
cessing) times, inner order of service and the sequence of service
2020-03-04 01:10. Page 1 of 1–8.
Unpublishedworkingdraft.
Notfordistribution.
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
ValueTools ’20, May 18-20, 2020, Tsukuba, Japan Yair Liberman and Uri Yechiali
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
stations in case of a tandem network. Studies on manufacturing
systems examined the effect of system parameters on productivity
efficiency and production costs. Regarding the single server model,
Levi and Yechiali (1975) [16] indicated the decomposition phenom-
enon of waiting times for the M/G/1 queue with multiple vaca-
tion, and derived the corresponding Laplace-Stieltjes Transform
(LST). Scholl and Kleinrock [23] investigated the multiple vacation
M/G/1 queue and derived the LSTs of waiting times under the FIFO,
LIFO and ROS (Random Order of Service) regimes. Rosenberg and
Yechiali (1993) [22] studied the batch arrivalMx /G/1 queue withsingle and with multiple vacations and derived explicit formulas
for the LSTs, means and second moments of the waiting time for
both FIFO and LIFO service-order regimes. They showed that al-
though E[WF I FO ] = E[WLI FO ], the second moment under LIFO
is larger than that under the FIFO, and gave an explicit formula,
E[(W
qF I FO )
2]= E
[(W
qLI FO )
2]/(1 − ρ), for all cases. The current
study relies on those results to compute the means and variances
of a product’s quality for theM/G/1 queue under both LIFO and
FIFO service regimes.
Research over a network of queues has began with the seminal
works of R.R.P. Jackson (1954 [12]; 1956 [13]) and J.R Jackson (1957
[10]; 1963 [11]). They developed a Markovian model, called Jackson
network, and characterized the transition probabilities and basic
system parameters of such networks. In a further work by Dallery
and Frein (1993) [7] a manufacturing system was investigated as an
open tandem Jackson network with either finite or infinite buffer
between sites. They showed how using the decomposition phenom-
ena allows the computation of basic system performance measures.
Other researches (see Yechiali, 1988 [27]; Brandon and Yechiali, 1991
[4]) studied the n-node tandem queueing network with Bernoulli
feedback to the first node and derived the LST of the total sojourn
time of a product in the system. Perlman and Yechiali (2019) [21]
investigated an n-site tandem network where each site consists of
two stages: a processing stage and an inspection stage. The pro-
cessing operation is a generally distributed random duration which
either does or does not conclude successfully; in the latter case,
the operation is repeated immediately. Once the processing stage
concludes successfully, the product goes through an inspection
stage which determines whether it will (i) require an additional
processing and move forward to the next station; or (ii) it is found
‘good’ and exits the system with quality value depending on its
sojourn time; or (iii) it is declared as ‘failed’and exits the system
with zero quality value. In their study they derived explicit results
for the mean sojourn time of a product in the system, and its mean
quality while assuming that each site is comprised of twoM/M/1-
type independent queues with FIFO service regime. In our work
we expand this model and give an explicit formula for a tandem
Jackson network where each site is an M/M/1 queue with LIFO
service regime, and compare the results between the FIFO and the
LIFO disciplines.
This paper treats two basic models (i) an M/G/1-type single-
server queue, and (ii) an open tandem Jackson network. The objec-
tive is to analyze the quality of products traversing through each
system under two service disciplines — FIFO and LIFO. In Section
2 theM/G/1 model is presented, and formulas are derived for the
quality’s mean and variance for both FIFO and LIFO service regimes.
Two service time distributions are investigated: Exponential, and
two-phase Erlang. The means and variances under each service
time distribution are compared. It is shown that FIFO is not neces-
sarily better then LIFO, as probably expected, but rather, the mean
quality under LIFO is better then the mean quality under FIFO for
all values of the traffic intensity. In Section 3 the tandem Jackson
network is analyzed and formulas are derived for the mean quality
of a product passing through the system, under both the FIFO and
LIFO service regimes. Numerical results further show the relations
between the two service disciplines. Conclusion are discussed in
Section 4.
2 MODEL FORMULATION - SINGLE SERVERWe consider an M/G/1 queueing system where a Poisson stream of
jobs (customers, calls, particles, products, etc.), with rate λ, flowsinto a single-server station. In steady state, let L denote the number
of jobs in the system, and let Pn = P(L = n), n = 0, 1, .... Service
time, B, of an arbitrary job is generally distributed with finite mean
E[B], finite second moment E[B2], and Laplace Stieltjes Transform
(LST) B(s). LetW qdenote the queueing time (not including service)
of a job, and letW =W q + B denote its overall sojourn time. Let
ρ = λE[B] < 1.
Consider the FIFO (First In First Out) service regime. The LST
of the queueing time of an arbitrary job is given by (see Kleinrock
(1975) [15], p. 199)
W q F I FO (s) =s(1 − ρ)
[s − λ + λB(s)](1)
which is known as the Pollaczek—Khinchine formula. The mean
queueing time, E[WqF I FO ], is given by
E[WqF I FO ] =
λE[B2]
2(1 − ρ). (2)
The LST of the sojourn time of a job is given by
WF I FO (s) = W q F I FO (s)B(s) =s(1 − ρ)B(s)
[s − λ + λB(s)]. (3)
Let θ denote the length of a busy period, which is the time from
the first arrival to an empty system until the system is empty again
for the first time. The mean and LST of θ are the solution for the
following equations (see Kleinrock (1975) [15], p. 212): E[θ ] = E[B]1−ρ ,
and
θ (s) = B[s + λ − λθ (s)]. (4)
Now, consider the non-preemptive LIFO (Last In Fast Out) service
discipline where the server serves each job completely, and imme-
diately afterwards begins to serve the last job to have arrived (if
any). The LST and mean of the queueing time of an arbitrary job
are given (see Rosenberg and Yechiali 1993 [22]) by
W qLI FO (s) = (1 − ρ) +λ[1 − θ (s)
]s + λ
[1 − θ (s)
] , (5)
and
E[WqLI FO ] =
λE[B2]
2(1 − ρ)= E[W
qF I FO ]. (6)
2020-03-04 01:10. Page 2 of 1–8.
Unpublishedworkingdraft.
Notfordistribution.
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
Quality-Dependent Stochastic Networks:Is FIFO Always Better Than LIFO? ValueTools ’20, May 18-20, 2020, Tsukuba, Japan
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
The LST of the sojourn timeWLI FO is given by
WLI FO (s) = WqLI FO (s)B(s) = (1 − ρ)B(s) +
λ[1 − θ (s)
]B(s)
s + λ[1 − θ (s)
] . (7)
It follows (see Rosenberg and Yechiali (1993)[22]), that although
E[WqF I FO ] = E[W
qLI FO ], the second moments are related by
E[(WqLI FO )
2] = 1
1−ρ E[(WqF I FO )
2].
2.1 Quality analysis forM/G/1 queueWhile waiting in line, a product suffers degradation in its quality.
The quality of a product that sojournsW units of time is considered
here as Q = δ e−γW , where δ is the potential quality in caseW =0, and γ is a positive constant serving as a scaling factor. This
formulation follows other works dealing with product degradation
(see, e.g. Perlaman and Yachiali (2019) [21] and references there).
For the M/G/1 queue, the expected value of a product’s quality
under the FIFO service regime is given by (see Eq. (3))
E[QF I FO ] = E[δe−γWF I FO ] = δWF I FO (γ ) = δγ (1 − ρ)B(γ )
γ − λ + λB(γ ). (8)
Clearly, when ρ → 1, E[WF I FO ] → ∞ and E[QF I FO ] → 0. How-
ever, when λ → 0 (ρ → 0), E[QF I FO ] → δ B(γ ). Indeed, inthe latter case, upon arrival, every product finds an empty sys-
tem so that its sojourn time is only its service duration B. Thus,
E[QF I FO ] = E[−δe−γ B ] = δ B(γ ).The second moment ofQF I FO can be calculated by using Eq. (8):
E[Q2
F I FO ] = E[δ2e−2γWF I FO ] =δ2WF I FO (2γ ) =
δ22γ (1 − ρ)B(2γ )
2γ − λ + λB(2γ ).
(9)
The Variance of QF I FO is calculated by
V [QF I FO ] =E[Q2
F I FO ] − E2[QF I FO ] =
δ22γ (1 − ρ)B(2γ )
2γ − λ + λB(2γ )−
[δγ (1 − ρ)B(γ )
γ − λ + λB(γ )
]2
.(10)
Now, when ρ → 1, V [QF I FO ] → 0, while when λ → 0 (ρ → 0),
V [QF I FO ] → δ2B(2γ ) − δ2B2(γ ). (11)
Regarding the LIFO service discipline, by using Eq. (7) the mean
quality of a product is calculated as
E[QLI FO ] =δE[e−γWLI FO ] = δWLI FO (γ ) =
δ
[(1 − ρ)B(γ ) +
λ[1 − θ (γ )]B(γ )
γ + λ[1 − θ (γ )
] ] . (12)
In this case too, limλ→0E[QLI FO ] = δ B(γ ). However, when ρ → 1
(i.e. λ → 1
E[B] ),
lim
ρ→1
E[QLI FO ] =[1 − θ (γ )]B(γ )
E[B]γ + [1 − θ (γ )]. (13)
The second moment of QLI FO is calculated as
E[Q2
LI FO ] =E[δ2e−2γWLI FO ] = δ2E[e−2γWLI FO ] = δ2WLI FO (2γ )
= δ2(1 − ρ)B(2γ ) + δ2λ[1 − θ (2γ )B(2γ )
2γ + λ[1 − θ (2γ )].
(14)
The Variance of QLI FO can be computed by using Eq.(12) and (14):
V [QLI FO ] =E[Q2
LI FO ] − E2[QLI FO ] =
δ2(1 − ρ)B(2γ ) + δ2λ[1 − θ (2γ )]B(2γ )
2γ + λ[1 − θ (2γ )]−
δ2[(1 − ρ)B(γ ) +
λ[1 − θ (γ )]B(γ )
γ + λ[1 − θ (γ )
] ]2.(15)
For the LIFO case, when ρ → 1 (i.e. λ → 1
E[B] ),
lim
ρ→1
V [QLI FO ] =
δ2 lim
ρ→1
[[1 − θ (2γ )]B(2γ )
E[B]2γ + [1 − θ (2γ )]−
([1 − θ (γ )]B(γ )
E[B]γ + [1 − θ (γ )]
)2].
(16)
For λ → 0, as in the FIFO case, from Eq. (15),
V [QLI FO ] → δ2B(2γ ) − δ2B2(γ ). (17)
In Sections 2.2, and 2.3 we consider two service distributions,
namely, (i) Exponential and (ii) two-phase Erlang.
2.2 Quality analysis for Exponential servicetimes
2.2.1 FIFO Service Regime. For theM(λ)/M(µ)/1 queue, where
B ∼ Exp(µ) and B(s) =µ
µ+s , the sojourn time of a product under
the FIFO service regime,WF I FO , is distributed Exponentially with
parameter µ − λ. Thus,
WF I FO (s) =(µ − λ)
(µ − λ) + s. (18)
Then, the expected quality of a product is given by Eqs. (8) and
(18):
E[QF I FO ] = δWF I FO (γ ) = δ(µ − λ)
(µ − λ) + γ. (19)
As in Section 2.1, when λ → 0, E[QF I FO ] → δµ
µ+γ = δ B(γ ),
whereas when ρ → 1 (i.e. λ → µ), E[QF I FO ] → 0.
The Variance of QF I FO is calculated by using Eq. (10):
V [QF I FO ] = δ2(µ − λ)
[1
µ + 2γ − λ−
µ − λ
(µ + γ − λ)2
]. (20)
It follows that, as in Section 2.1, when λ → µ (i.e. ρ → 1),V [QF I FO ] →
0 (see also Fig. 1), while when λ → 0,
V [QF I FO ] → δ2µ
[1
µ + 2γ−
µ
(µ + γ )2
]= δ2
[B(2γ ) − B(γ )2
]. (21)
2020-03-04 01:10. Page 3 of 1–8.
Unpublishedworkingdraft.
Notfordistribution.
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
ValueTools ’20, May 18-20, 2020, Tsukuba, Japan Yair Liberman and Uri Yechiali
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
2.2.2 LIFO Service Regime. For the LIFO service discipline in an
M(λ)/M(µ)/1 queue, where B(γ ) =µ
µ+γ , we get by Eq. (12):
E[QLI FO ] =δ (µ − λ)
µ + γ+
δλµ[1 − θ (γ )]
(µ + γ )(γ + λ[1 − θ (γ )]). (22)
Note that the length of the busy period in a work conserving system
is independent of the service regime. Thus, by using Eq. (4), we
obtain
θ (γ ) = B(γ + λ − λθ (γ )) =µ
µ + [γ + λ − λθ (γ )],
(23)
resulting in
θ (γ ) =γ + λ + µ −
√(γ + λ + µ)2 − 4λµ
2λ. (24)
By substituting Eq. (24) in Eq. (22) we get an explicit expression for
E[QLI FO ]. That is,
E[QLI FO ] =
δ
µ + γ
[2µ − λ −
2γ µ
λ + γ − µ +√(γ + λ + µ)2 − 4λµ
].
(25)
Indeed, as in Section 2.1, when λ → 0, E[QLI FO ] = δµ
µ+γ =
δ B(γ ), whereas when ρ → 1 (i.e. λ → µ)
E[QLI FO ] →δ
µ + γ
[µ −
2γ µ
γ +√γ 2 + 4γ µ
]. (26)
The variance of QLI FO , when substituting B(2γ ) =µ
µ+2γ in Eq.
(15), results in
V [QLI FO ] = δ2(µ − λ)
(µ + 2γ )+ δ2
λµ[1 − θ (2γ )]
(µ + 2γ )(2γ + λ[1 − θ (2γ )])
−δ2
(µ + γ )2
[(µ − λ)2 +
2λµ(µ − λ)[1 − θ (γ )]
γ + λ[1 − θ (γ )]+[
λµ[1 − θ (γ )]
γ + λ[1 − θ (γ )]
]2],
(27)
where
θ (2γ ) =2γ + λ + µ −
√(2γ + λ + µ)2 − 4λµ
2λ. (28)
2.2.3 Comparison between LIFO and FIFO forM(λ)/M(µ)/1queue. In this section the mean qualities and quality variances
under the LIFO and under the FIFO service regimes are compared
for theM/M/1 queue. Using Eqs. (19) and (25), the ratio between
the two mean qualities is given by
E[QLI FO ]
E[QF I FO ]=
(µ − λ + γ )
(µ − λ)(µ + γ )
[2µ − λ −
2γ µ
λ + γ − µ +√(γ + λ + µ)2 − 4λµ
].(29)
Propositions 1 and 2 below show the advantage of the LIFO over
the FIFO with respect to mean qualities.
Proposition 2.1. 1 For µ,γ > 0 and 0 < λ ≤ µ,
limλ→µE[QLI FO ]
E[QF I FO ]= ∞.
Proof. The claim follows directly from Eq. (29). □
Proposition 2.2. 2 For γ > 0 and 0 < λ < µ,E[QLI FO ]
E[QF I FO ]> 1.
Proof. By Eq. (29)E[QLI FO ]
E[QF I FO ]> 1 ⇐⇒
(µ − λ + γ )(2µ − λ)[λ + γ − µ +
√(γ + λ + µ)2 − 4λµ
]+ 2γ µ >
(µ − λ)(µ + γ )[λ + γ − µ +
√(γ + λ + µ)2 − 4λµ
].
(30)
LetC B λ+γ−µ+√(γ + λ + µ)2 − 4λµ denote the the numerator
of θ (γ ) in Eq. (24). Clearly,C > 0. SubstitutingC in Eq. (30) translates
to
(µ − λ)(2µ − λ)C + γ (2µ − λ)C + 2γ µ > (µ − λ)(µ + γ )C ⇐⇒
(µ − λ)2C + γ µC + 2γ µ > 0,
(31)
which holds since γ µ > 0 and (µ − λ)2 > 0. □
Figure 1 depicts the behaviour of the mean qualities (Figure 1(a))
and variances (Figure 1(b)) for both FIFO (red) and LIFO (blue)
service regimes as functions of λ when µ = 20, γ = 1, δ = 1.
E[QLI FO ], is a monotonously decreasing concave function of λ,approaching 0.76 as λ → µ (see Eq. (26)). E[QF I FO ] is also concave,
but in contrast to E[QLI FO ], it decreases sharply when λ gets closerto µ, reaching 0 as λ = µ. Figure 1(a) illustrates graphically the
claims of proposition 1 and 2.
(a) E[Q ] as a function of λ (b) V [Q ] as a function of λ
Figure 1: Mean and variance of a product’s quality under FIFO andunder LIFO service regimes for Exponential service times as a
function of λ, when µ = 20, γ = 1, δ = 1
Figure 1(b) depicts the behaviour of the variances of the two
service regimes as functions of λ for the same parameter values as in
figure 1(a). It is seen that both variances increase when λ is small or
moderate. As long as λ is smaller then 18.2,V (QF I FO ) < V (QLI FO ),
but when λ exceeds 18.2, the ratio turns over and V (QF I FO ) >
V (QLI FO ). As λ further increasesV (QF I FO ) changes its direction at
its highest valueV (QF I FO ) = 0.09, and then drops sharply towards
0, crossing V (QLI FO ) at λ = 19.6.
The above observations are explained as follows: When λ is
small, there is a significant probability under the LIFO regime that
the last arriving product will be admitted to service before another
arrival occurs, while if not, the product will have to wait at least
one busy period. These two opposite possibilities increase the value
of the quality’s variance under FIFO regime. When λ approaches µ,the queue becomes very large and every product waits a very long
time, so that V [QF I FO ] decreases sharply.
2020-03-04 01:10. Page 4 of 1–8.
Unpublishedworkingdraft.
Notfordistribution.
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
Quality-Dependent Stochastic Networks:Is FIFO Always Better Than LIFO? ValueTools ’20, May 18-20, 2020, Tsukuba, Japan
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
Figures 2 further depicts the ratios between the mean qualities of
the two regimes and the ratio between their variances. In figure 2(a)
it is seen that the ratio increases slowly up to λ = 18 and then jumps
to ∞. Figure 2(b) illustrates that the ratio between the variances
increases up to λ = 13.5, then decreases sharply up to λ = 19.2, and
then jumps almost instantly towards ∞. These turning points in
the ration coincides with the two crossing points of Figure 1(b).
(a) Ratio between Expectedquality in LIFO regime vs
FIFO regime(b) Ratio between the qualityVariances of the two regimes
Figure 2: Comparing quality means and variances for FIFO andLIFO service regimes for Exponential service times, when µ = 20,
γ = 1, δ = 1
2.3 Quality analysis for two-phase Erlangservice times
In this section we consider anM/G/1 queue, with Poisson arrival
rate λ and Erlang service time having two consecutive exponential
(µ) durations. That is, B ∼ Eralnд(2, µ). We have B(s) =( µµ+s
)2
,
ρ = 2λµ . The stability condition ρ < 1 translates here to λ <
µ2.
2.3.1 FIFO Service Regime inM(λ)/E(2, µ)/1 queue. Underthe FIFO regime, by using Eq. (8), the mean quality of a departing
product is given by
E[QF I FO ] = δµ(µ − 2λ)
(µ + γ )2 − λ(2µ + γ ). (32)
By using Eq. (10), the variance of QF I FO is given by
V [QF I FO ] = δ22γ µ(µ − 2λ)
(2γ − λ)(µ + 2γ )2 + λµ2−
[δ
γ µ(µ − 2λ)
(γ − λ)(µ + γ )2 + λµ2
]2
.
(33)
2.3.2 LIFO Service Regime inM(λ)/E(2, µ)/1 queue. Under LIFO,by using Eq. (12) the mean quality is given by
E[QLI FO ] = δµ(µ − 2λ)
(µ + γ )2+ δ
λµ2[1 − θ (γ )]
(µ + γ )2(γ + λ[1 − Q(γ )]), (34)
where here θ (γ ) can be calculated by Eq. (4):
θ (γ ) = B(γ + λ − λθ (γ )) =µ2[
µ +[γ + λ − λθ (γ )
] ]2 (35)
This leads to a cubic equation in θ (γ ):
λ2θ3(γ ) − 2λ(µ + γ + λ)θ2(γ ) + (µ + γ + λ)2θ (γ ) − µ2 = 0. (36)
Denote the coefficients in the 3rd degree polynomial given in Eq.
(36) as a = λ2, b = −2λ(µ + γ + λ), c = (µ + γ + λ)2, and d = −µ2.
Then, by using Cardano’s Formula for solving a cubic equation (see
Witula and Slota, 2009 [26]) we get three roots, only one of them is
real and positive, so it is the only valid candidate for θ (γ ):
θ (γ ) = S +T +2(γ + µ + λ)
3λ, (37)
where
S =3
√R +
√P3 + R2 T =
3
√R −
√P3 + R2 , (38)
while,
R =9abc − 27a2d − 2b3
54a3=
2(γ + µ + λ)3 + 27λµ2
54λ3
P =3ac − b2
9a2= −
(γ + µ + λ)2
9λ2
(39)
E[QLI FO ] is calculated for the M(λ)/E(2, µ)/1 queue by substi-
tuting first Eq. (39) in Eq. (38), then Eq. (38) in Eq (37), and finally
Eq. (37) in Eq. (34).
The Variance of QLI FO is given by
V [QLI FO ] =δ2µ(µ − 2λ)
(µ + 2γ )2+ δ2
λµ2[1 − θ (2γ )]
(µ + 2γ )2(2γ + λ[1 − θ (2γ )])−
δ2µ2(µ − 2λ)2
(µ + γ )4+ 2δ2
λµ3(µ − 2λ)[1 − θ (γ )]
(µ + γ )4(γ + λ[1 − θ (2γ )])−
δ2λ2µ4[1 − θ (γ )]2
(µ + γ )4(γ + λ[1 − θ (2γ )])2,
(40)
where θ (2γ ) is the real positive root of the following cubic equation:
λ2θ3(2γ ) − 2λ(µ + 2γ + λ)θ2(2γ ) + (µ + 2γ + λ)2θ (2γ ) − µ2 = 0
(41)
which is solved by Cardano’s formula.
2.3.3 Comparison between LIFO and FIFO inM(λ)/E(2, µ)/1queue. In this sub-section the FIFO and LIFO service regimes are
compared numerically with respect to theM/E2/1 queue.Figures 3 bellow depicts the results for the mean qualities (Figure
3(a)) and variances (Figure 3(b)) for both FIFO (red) and LIFO (blue)
regimes as functions of λ, when µ = 20, γ = 1, δ = 1. E[QLI FO ] is a
concave monotonic decreasing function of λ, and approaches 0.7 asρ → 1, i.e., λ →
µ2= 10. In comparison, E[QF I FO ] is also concave,
but decreases sharply when ρ gets closer to 1, reaching 0 at λ = 10.
Note that E[QLI FO ]/E[QF I FO ] > 1 for all λ.
(a) E[Q ] as a function of λ (b) V [Q ] as a function of λ
Figure 3: Mean and variance of a product’s quality under FIFO andunder LIFO service regimes for two-phase Erlang service times
(E(µ , 2)), as functions of λ, when µ = 20, γ = 1, δ = 12020-03-04 01:10. Page 5 of 1–8.
Unpublishedworkingdraft.
Notfordistribution.
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
ValueTools ’20, May 18-20, 2020, Tsukuba, Japan Yair Liberman and Uri Yechiali
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
Figure 3(b) depicts the behaviour of the variances of the two
service regimes as a function of λ for the same parameter values as
in Figure 3(a). It is seen that both variances increase when λ is small
or moderate. When λ is smaller than 8.8, V [QF I FO ] < V [QLI FO ],
but when λ exceeds λ = 8.8, the ratio turns over and V [QLI FO ] <
V [QF I FO ]. V [QF I FO ] reverses its direction at its highest value
V [QF I FO ] = 0.084 and drops sharply towards 0, crossingV [QLI FO ]
at λ = 9.7.
Figure 4 depicts the ratios between the mean qualities of the two
regimes and the ratio between their variances. In Figure 4(a) the
ratio between the mean qualities increases slowly up to λ = 8.5
and then jumps rapidly to∞. Figure 4(b) illustrates that the ratio
between the variances increases up to λ = 6.3, then decreases
sharply up to λ = 9.4 and finally jumps almost instantly towards
∞. Observe that Figures 4(a) and 4(b) are similar, respectively, to
Figures 2(a) and 2(b).
(a) Ratio between meanqualities:
E[QLI FO ]/E[QF I FO ]
(b) Ratio between the qualityVariances:
E[QLI FO ]/E[QF I FO ]
Figure 4: Comparing quality’s means and variances for FIFO andLIFO service regime for M (λ)/E(2, µ)/1 queue, as a function of λ,
when µ = 20, γ = 1, δ = 1
3 TANDEM JACKSON NETWORK (TJN) WITHTIME-DEPENDENT QUALITYDETERIORATION
Consider an n-site stochastic tandem Jackson network (TJN) where
products arrive to the first site at a Poisson rate λ, and then, after
being processed and inspected at each site, move uni-directionally
from site to site, as depicted in Figure 5 (see Perlman and Yechiali,
2019 [21]).
Figure 5: Tandem Stochastic Network. "Proc(i)" ("Ins(i)") indicatesthe processing (inspection) stage of site i
Each site consists of two separate and independent stages: a
processing stage, followed by an inspection stage, where each stage
is anM/M/1-type queue. A single processing attempt at stage i, Bi ,
is exponentially distributed with mean 1/µi and LST Bi (s) =µi
s+µi ,
and it concludes either successfully, with probability ri , or fails withthe complementary probability 1 − ri . In the latter case, the pro-
cessing operation is immediately repeated with the same success
probability, until a successful attempt. Thus, the total processing
time of a product at site i , Gi , is a Geometric sum of iid Exponen-
tial random variables, and is distributed Exponentially with mean
E(Gi ) =1
µi ri and LST given by G(s) =ri µi
s+ri µi . It is well known (see
Takagi, 1991 [24]) that the output process from an M(λ)/M(∗)/1
queue is also Poisson with the same rate λ, and that the inter-
departure times form a renewal process. Thus, each site can be
treated as an isolatedM/M/1 queue. Upon successful processing
completion at site i, the product moves to the inspection stage of
this site. The inspection time, Di , is exponentially distributed with
mean 1/ξi and LST Di (s) =ξi
s+ξi(i = 1, 2, ...,n).
Upon conclusion of the inspection, the state of the product is
determined according to the following possibilities: (i) with proba-
bility ai the product advances to queue i+1 (1 ≤ i ≤ n−1); (ii) with
probability pi the product is declared to be ’good’ and exits the sys-
tem with a quality value Q depending on the total time it sojourned
in the system; and (iii) with probability fi the product is declaredas ’failed’, discarded and exits the system with quality 0. Clearly,
ai + pi + fi = 1 for every i = 1, 2, ...,n. Thus, the arrival rate to the
processing stage at site i is λi = λ∏i−1
k=1 ak for 1 < i ≤ n− 1, where∏0
k=1 ak = 1. Out of site n, an = 0, and pn + f n = 1. The traffic load
condition for the network to reach stability is λi < min{ri µi , ξi }for each site 1 ≤ i ≤ n.
During its traversal time through the system the product suf-
fers degradation in its quality, depending on the total time it has
sojourned in the system. Let Tj denote a product’s accumulated
traversal time through sites 1 to j . If the product exits the system at
site j (1 ≤ j ≤ n), its quality is either Q = δe−γTj with probability
pj , or zero with probability fj . As indicated above, both stages are
M/M/1 queues. LetWi and Vi denote the total sojourn time in the
processing stage, and in the inspection stage at site i , respectively.Then, the total traversal time of a product through sites 1 to j is
Tj =∑ji=1(Wi +Vi ).
3.1 FIFO Service RegimeThe total sojourn time of a product in an M(λ)/M(µ)/1 queue is
exponentially distributed with mean1
µ−λ . Thus, since the total
processing time in site i is exponentially distributed with parameter
ri µi , and the arrival rate is λi , under the FIFO regime,W F I FOi (a
slight change of notation), the product’s total processing time at
site i , is exponentially distributed with parameter ri µi −λi and LST
W F I FOi (s) =
ri µi−λiri µi−λi+s
.
Similarly, the inspection stage at site i is also an M/M/1 queue
with arrival rate λi and inspection rate ξi . Consequently, the so-journ time V F I FO
i of a product in the inspecting stage there is
also exponentially distributed with parameter ξi − λi and LST
V F I FOi (s) =
ξi−λiξi−λi+s
.
The mean traversal time in the system E[T F I FO ] is given by (see
Perlman and Yechiali [2019])
E[T F I FO ] =
n∑j=1
( ( j−1∏k=1
ak)(pj + fj )
j∑i=1
( 1
ri µi − λi+
1
ξi − λi
) ).
(42)
2020-03-04 01:10. Page 6 of 1–8.
Unpublishedworkingdraft.
Notfordistribution.
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
Quality-Dependent Stochastic Networks:Is FIFO Always Better Than LIFO? ValueTools ’20, May 18-20, 2020, Tsukuba, Japan
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
The mean traversal time for a product exiting the system as ’good’
is given by (Perlman and Yechiali, 2019 [21])
E[T F I FO |дood] =
∑nj=1
( (∏j−1k=1 ak
)pj
∑ji=1
(1
ri µi−λi+ 1
ξi−λi
) )∑nj=1
( ∏j−1k=1 ak
)pj
(43)
and the mean quality of a product leaving the system is given by
E[QF I FO ] =
n∑j=1
( ( j−1∏k=1
ak)pjδ
j∏i=1
( ri µi − λiri µi − λi + γ
·ξi − λi
ξi − λi + γ
) ),
(44)
where λi = λ∏i−1
k=1 ak .
3.2 LIFO Service RegimeConsidering site i , the LST ofW LI FO
i is given by Eq. (7). Thus, with
ρ = λiri µi and Gi (s) =
ri µiri µi+s , the LST of the the sojourn time in the
processing stage is given by
W LI FOi (s) =
ri µi − λiri µi + s
+λiri µi
[1 − θi (s)
](ri µi + s)(s + λi
[1 − θi (s)
]), (45)
where the LST of θi , the corresponding busy period, is given by
θi (s) =s + λi + ri µi −
√(s + λi + ri µi )2 − 4λiri µ
2λi. (46)
Similarly, the LST of the inspection time in site i , V LI FOi , is
calculated with the aid of Eq. (7). Thus, with ρ = λiξi
and Di (s) =
ξiξi+s
, the LST of V LI FOi is given by
V LI FOi (s) =
ξi − λiξi + s
+λiξi
[1 − ϕi (s)
](ξi + s)(s + λ
[1 − ϕi (s)
]), (47)
where ϕi (s), denoting the busy period in the inspection stage at
site i , is given by
ϕi (s) =s + λi + ξi −
√(s + λi + ξi )2 − 4λiξi2λi
. (48)
Thus, the mean quality of an arbitrary product exiting the system
is given by:
E[QLI FO ] =
n∑i=1
( i−1∏k=1
ak)piE[Q
LI FOi ], (49)
where E[QLI FOi ], the mean quality of a product exiting the system
in site i, is given by:
E[QLI FOi ] = E[−γT
LI FOi ] = δT LI FO
i (γ ) = δi∏
k=1
W LI FOi (γ ) · V LI FO
i (γ )
= δi∏
k=1
(rk µk − λkrk µk + γ
+λkrk µk
[1 − θk (γ )
](rk µk + γ )(γ + λk
[1 − θi (γ )
])
)×(
ξk − λkξk + γ
+λk ξk
[1 − ϕk (γ )
](ξk + γ )(γ + λk
[1 − ϕk (γ )
])
).
(50)
3.3 Comparison between LIFO and FIFO in TJNIn the following section the FIFO and LIFO service regimes are
compared numerically for the tandem Jackson network with n = 2
sites. Two cases are investigated with various parameter values: (i)
identical sites, and (ii) non-identical sites.
3.3.1 Identical sites (n=2). Assume that all processing stages are
statistically identical and all inspection stages are also identical and
compare mean quality as a function of λ. Figure 6 illustrates theresults of E[Q] for various values of ri while γ = 1, δ = 1, µi = 20,
ξi = 20. It is seen that. for all values of ri , LIFO is better than FIFO
for all values of λ. Under both service regimes, as ri increases, themean quality increases due to the decrease in sojourn time at every
station. In addition, for ri = 0.1 (Figure 6a) the shape of E[QLI FO ]
is approximately linear while for ri = 0.5 and ri = 1 E[QLI FO ] is
concave. In all cases E[Q] decreases when λ increases.
(a) E[Q ] for ri = 0.1 (b) E[Q ] for ri = 0.5 (c) E[Q ] for ri = 1
Figure 6: Mean product quality (FIFO and LIFO) in TJN with twonodes as a function of λ for various values of ri while γ = 1, δ = 1,
µi = 20, ξi = 20
Figure 7 illustrates the results of E[Q] as a function of λ for
different values of the shape parameter γ , while δ = 1, ri = 1,
µi = 20, ξi = 20. It is seen that for all values of γ , LIFO is better
than FIFO for all values of λ. When γ = 0.5, for both FIFO and LIFO,
the mean quality is a concave function with increasing difference
between FIFO and LIFO as λ approaches µ. In comparison, forγ = 25
and γ = 50, the mean quality under both regimes is convex, and the
difference between FIFO and LIFO decreases as γ increases. This is
due to the increasing deterioration rate as γ increases.
(a) E[Q ] for γ = 0.5 asa function of λ
(b) E[Q ] for γ = 25 as afunction of λ
(c) E[Q ] forγ = 50 as afunction of λ
Figure 7: Mean product quality for TJN (FIFO and LIFO) with twonodes for various values of γ while δ = 1, ri = 1, µi = 20, ξi = 20
3.3.2 Non-identical processing sites. The parameters for the
first site are: r1 = 1, p1 = 0, a1 = 0.8 and f1 = 0.2, while the
parameters for the second site are: r2 = 1, p2 = 0.2, and f2 = 0.8.
Consider two cases: (i) µ1 = 30, µ2 = 60 and (ii) µ1 = 60, µ2 = 30.
For simplicity we assume that inspection stages are identical with
ξ1 = ξ2 = 20. Figure 8 bellow depicts the value for E[Q] whenδ = 1, and γ = 1. Note that when the processing rate at site 1 is
higher (Fig 8(a)), the deterioration as ρ approaches 1 in both service
2020-03-04 01:10. Page 7 of 1–8.
Unpublishedworkingdraft.
Notfordistribution.
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
ValueTools ’20, May 18-20, 2020, Tsukuba, Japan Yair Liberman and Uri Yechiali
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
regimes is almost identical, while when the processing rate at site 1
is lower (Fig 8(b)), E[QF I FO ] is bounded (by 0.12) while E[QF I FO ]
is decreasing sharply towards 0, as ρ approaches 1. Those results areexplained by examining the results in previous sections regarding
the M/G/1 queue: when ρ approaches 1, E[QF I FO ] deteriorates
rapidly towards 0.
(a) E[Q ] for µ1 = 60, µ2 = 30 asa function of λ
(b) E[Q ] for µ1 = 30, µ2 = 60 asa function of λ
Figure 8: Mean product quality in a two-node TJN (FIFO and LIFO)with non identical processing stages, while δ = 1, ri = 1, γ = 1,
ξi = 120
4 CONCLUSIONSThis paper analyses the quality of a product traversing through (i) a
single serverM/G/1-type queue, and (ii) a tandem Jackson network.
For theM/G/1 queue, two service time distributions are specifically
studied - Exponential and two-phase Erlang (other service time
distributions are discussed in a technical report by Liberman and
Yechiali, 2019 [17]).
Assuming that the quality of a product deteriorates exponentially
with the time it sojourns in the system, two service disciplines
are compared - FIFO vs. LIFO (the investigation of other service
regimes such as Random Order of Service or Processor Sharing
are left to a further research). Interestingly, it is shown that FIFO
is not necessarily better then LIFO. Numerical results show the
behaviour of the quality mean, as well as the quality variance, as
functions of the various system parameters. It is shown that the
mean quality under LIFO is always greater then FIFO. The first
is bounded below when the traffic intensity approaches 1, while
the latter drops sharply to zero. Similar qualitative results hold
for a tandem Jackson network. A further possible research may
look at concave deteriorating quality function where the product is
considered as perfectly good up to some fixed time threshold.
REFERENCES[1] Tal Avinadav and Teijo Arponen. 2009. An eoq model for items with a fixed
shelf-life and a declining demand rate based on time-to-expiry technical note.
Asia-Pacific Journal of Operational Research 26, 06 (Dec. 2009), 759–767. https:
//doi.org/10.1142/S0217595909002456
[2] Emre Berk and Ãlkü Gürler. 2008. Analysis of the (Q, r) Inventory Model for
Perishables with Positive Lead Times and Lost Sales. Operations Research 56, 5
(Oct. 2008), 1238–1246. https://doi.org/10.1287/opre.1080.0582
[3] Marco Bortolini, Maurizio Faccio, Emilio Ferrari, Mauro Gamberi, and Francesco
Pilati. 2016. Fresh food sustainable distribution: cost, delivery time and carbon
footprint three-objective optimization. Journal of Food Engineering 174 (April
2016), 56–67. https://doi.org/10.1016/j.jfoodeng.2015.11.014
[4] Jonathan Brandon and Uri Yechiali. 1991. A tandem Jackson network with
feedback to the first node. Queueing Systems 9, 4 (Dec. 1991), 337–351. https:
//doi.org/10.1007/BF01159221
[5] Marcello Colledani, Andras Horvath, and Alessio Angius. 2015. Production
quality performance in manufacturing systems processing deteriorating products.
CIRP Annals 64, 1 (Jan. 2015), 431–434. https://doi.org/10.1016/j.cirp.2015.04.122
[6] William L. Cooper. 2001. Pathwise Properties and Performance Bounds for a
Perishable Inventory System. Operations Research 49, 3 (June 2001), 455–466.
https://doi.org/10.1287/opre.49.3.455.11216
[7] Yves Dallery and Yannick Frein. 1993. On Decomposition Methods for Tandem
Queueing Networks with Blocking. Operations Research 41, 2 (April 1993), 386–
399. https://doi.org/10.1287/opre.41.2.386
[8] S. K. Goyal and B. C. Giri. 2001. Recent trends in modeling of deteriorating
inventory. European Journal of Operational Research 134, 1 (Oct. 2001), 1–16.
https://doi.org/10.1016/S0377-2217(00)00248-4
[9] Robert R. Inman, Dennis E. Blumenfeld, Ningjian Huang, Jingshan Li, and Jing
Li. 2013. Survey of recent advances on the interface between production system
design and quality. IIE Transactions 45, 6 (June 2013), 557–574. https://doi.org/
10.1080/0740817X.2012.757680
[10] James R. Jackson. 1957. Networks of Waiting Lines. Operations Research 5, 4
(Aug. 1957), 518–521. https://doi.org/10.1287/opre.5.4.518
[11] James R. Jackson. 1963. Jobshop-Like Queueing Systems. Management Science10, 1 (Oct. 1963), 131–142. https://doi.org/10.1287/mnsc.10.1.131
[12] R. R. P. Jackson. 1954. Queueing Systems with Phase Type Service. Journal of theOperational Research Society 5, 4 (Dec. 1954), 109–120. https://doi.org/10.1057/
jors.1954.23
[13] R. R. P. Jackson. 1956. Random Queueing Processes with Phase-Type Service.
Journal of the Royal Statistical Society: Series B (Methodological) 18, 1 (1956),
129–132. https://doi.org/10.1111/j.2517-6161.1956.tb00218.x
[14] Jongyoon Kim and Stanley B. Gershwin. 2005. Integrated quality and quantity
modeling of a production line. OR Spectrum 27, 2 (June 2005), 287–314. https:
//doi.org/10.1007/s00291-005-0202-1
[15] Leonard Kleinrock. 1975. Queueing systems, volume I: theory. John Wiley & Sons
New York.
[16] Y. Levi and U. Yechiali. 1975. Utilization of idle time in an M/G/1 queuing system.
Manag. Sci 22, 2 (1975), 202–211.[17] Yair Liberman and Uri Yechiali. 2019. Quality-Dependent Stochastic Networks:
Is FIFO Always Better Than LIFO? Master’s thesis. Department of Statistics and
Operations Research, Tel Aviv University, Tel Aviv, Israel.
[18] George Liberopoulos, George Kozanidis, and Panagiotis Tsarouhas. 2007. Perfor-
mance Evaluation of an Automatic Transfer Line with WIP Scrapping During
Long Failures. Manufacturing & Service Operations Management 9, 1 (Jan. 2007),62–83. https://doi.org/10.1287/msom.1060.0118
[19] George Liberopoulos, George Kozanidis, and Panagiotis Tsarouhas. 2007. Perfor-
mance Evaluation of an Automatic Transfer Line with WIP Scrapping During
Long Failures. Manufacturing & Service Operations Management 9, 1 (Jan. 2007),62–83. https://doi.org/10.1287/msom.1060.0118
[20] Raed Naebulharam and Liang Zhang. 2013. Performance Analysis of Serial
Production Lines with Deteriorating Product Quality. IFAC Proceedings Volumes46, 9 (Jan. 2013), 501–506. https://doi.org/10.3182/20130619-3-RU-3018.00105
[21] Yael Perlman and Uri Yechiali. 2019. On tandem stochastic networks with time-
deteriorating product quality. International Journal of Production Research 0, 0
(July 2019), 1–9. https://doi.org/10.1080/00207543.2019.1637034
[22] Eitan Rosenberg and Uri Yechiali. 1993. The Mx/G/1 queue with single and
multiple vacations under the LIFO service regime. Operations Research Letters 14,3 (Oct. 1993), 171–179. https://doi.org/10.1016/0167-6377(93)90029-G
[23] Michel Scholl and Leonard Kleinrock. 1983. On the M/G/1 Queue with Rest
Periods and Certain Service-Independent Queueing Disciplines. OperationsResearch 31, 4 (Aug. 1983), 705–719. https://doi.org/10.1287/opre.31.4.705
[24] H. TAKAGI. 1991. Queueing Analysis : A Foundation of Performance Evaluation.
Vacation and Priority Systems 1, 1 (1991). https://ci.nii.ac.jp/naid/10012612528/
[25] Adriaan Van Horenbeek, Jasmine Buré, Dirk Cattrysse, Liliane Pintelon, and
Pieter Vansteenwegen. 2013. Joint maintenance and inventory optimization
systems: A review. International Journal of Production Economics 143, 2 (June2013), 499–508. https://doi.org/10.1016/j.ijpe.2012.04.001
[26] Roman WituÅa and Damian SÅota. 2010. Cardano’s formula, square roots,
Chebyshev polynomials and radicals. J. Math. Anal. Appl. 363, 2 (March 2010),
639–647. https://doi.org/10.1016/j.jmaa.2009.09.056
[27] Uri Yechiali. 1988. Sequencing an N-Stage Process with Feedback. Probabilityin the Engineering and Informational Sciences 2, 2 (April 1988), 263–265. https:
//doi.org/10.1017/S0269964800000784
.
2020-03-04 01:10. Page 8 of 1–8.